Key facts
- Google's TPUs will be available on Oracle's cloud platform starting in 2025.
- OpenAI plans to purchase TPUs from Google and optimize its models for them.
- Mizuho forecasts Google's TPU revenue to reach $100 billion by 2028.
- Anthropic plans to spend $114.5 billion on Google compute between April 2026 and July 2033.
Google's AI chip business, centered around its Tensor Processing Units (TPUs), is increasingly posing a challenge to Nvidia's market leadership. A new deal with Oracle will make Google's TPUs available to Oracle cloud customers starting next year, marking the first time they will be offered on a major hyperscaler's platform outside of Google's own infrastructure. OpenAI is also reportedly planning to purchase TPUs and optimize its AI models for them, indicating a growing adoption of a multi-chip strategy by key AI developers. This diversification could potentially impact Nvidia's pricing power, as companies seek to hedge their bets and reduce reliance on a single chip provider. Google has been investing in its custom AI silicon for over a decade, initially to enhance its search products, and these chips are now poised for significant growth.